Switching oscillation-based fractional-domain high-sensitive diagnosis of machine turn insulation degradation

Stator turn insulation degradation is a significant cause of failure in electric machines fed by high-frequency switching power electronic devices, making early-stage detection challenging due to weak fault characteristic signatures. This paper presents a novel non-invasive monitoring framework that integrates a proposed low-to-high-frequency parametric grey-box model (GBM) of the motor with advanced signal processing techniques. Key electrical parameters of the motor are extracted using optimal fitting with impedance spectroscopy, which represent frequency-dependent behaviors influenced by insulation degradation. To enhance detection sensitivity, the method employs an optimized fractional Fourier transform at the optimal order, coupled with a customized resonance-centred Mel filter bank (FRFT-RMFB). Experimental results on a 0.75 kW induction motor with controlled insulation degradation, implemented by parallel capacitive loading, show a higher improvement in early-stage fault sensitivity. The proposed FRFT‑RMFB method enhances the sensitivity of turn insulation state estimation by nearly sevenfold compared with the conventional Fourier transform approach. The proposed framework is lightweight, generalized, and well-suited for real-time predictive maintenance, effectively minimizing false alarms in safety–critical applications such as industrial drives, electric vehicles, and high-speed rail systems. This work contributes to a new direction in frequency-domain diagnostics, combining accurate parametric modeling with high-resolution spectral feature extraction for early detection of insulation faults.